The CV Is Becoming Obsolete
AI is making the CV better as marketing and worse as information. The document may survive. Its status as the canonical record of a career should not.
For decades, the CV has compressed a career into one or two pages and asked an employer to infer future capability from employers, titles, dates and selected achievements. That arrangement made sense when professional information was expensive to collect, store and interpret.
The CV is fundamentally the wrong data structure for representing human capital.
The CV was an information-compression technology.
A career contains projects, decisions, transactions, clients, teams, problems, outcomes, failures, relationships, subject-matter knowledge and thousands of individual pieces of work. A traditional hiring process could not economically capture or interpret all of it.
The CV solved that problem through compression. A recruiter could scan twenty years in seconds. An organisation could store thousands of records. A candidate could distribute the same document repeatedly.
The trade-off was enormous information loss. A title such as Vice President, Investment Banking might represent hundreds of negotiations, analyses and decisions. A bullet stating that somebody “led strategic initiatives across European markets” might conceal years of work across very different countries, industries and operating problems.
The employer does not observe the underlying human capital. It observes a compressed representation of it. The rest is inference.
Hiring became an exercise in interpreting proxies.
Institutional evidencePrestigious employer becomes a proxy for quality. Title becomes a proxy for responsibility. Tenure becomes a proxy for capability. Education becomes a proxy for intelligence or trainability. Career progression becomes a proxy for performance.
Those signals are not meaningless. In standardised labour markets, a hiring manager may understand an employer's training model well enough for title and tenure to carry substantial information.
The weakness emerges when careers cross functions, industries, business models or geographies. The OECD describes work experience and university degrees as the primary proxies still used to assess skills, while noting both the difficulty of validating capability at scale and the relatively weak predictive power historically associated with years of experience and education.1
The problem is not merely missing data. It is interpretive burden. Somebody must reconstruct the person from the document.
The CV does not contain the career. It contains the fragments selected for the last transaction.
The document discards evidence before anyone knows what will matter.
A professional may spend four years in a role and allocate five bullets to it. Those bullets are expected to explain what the person did, how well they did it, what differentiated them from colleagues and why any of it matters for the next job.
Compression also forces selection. Candidates decide today which experiences deserve preservation, without knowing what may become relevant later. An unusual cross-border project, difficult stakeholder negotiation, operating failure or internal initiative can disappear because it did not support the immediate application.
The evidence existed. The document discarded it.
Human capital is the accumulated stock of knowledge, capability, experience, relationships and demonstrated behaviour developed over a career. The CV is only one temporary representation of that stock.
AI perfects the marketing document while weakening the signal.
Generative AI can improve weak language, align terminology with a vacancy, reorder experience and produce hundreds of role-specific variations cheaply. From the candidate's perspective, that is rational.
At market level, it compresses variance. If most candidates can produce professionally written, keyword-aligned CVs, writing quality and tailoring become less informative. A mediocre candidate with excellent assistance and an exceptional candidate using the same tools can produce superficially similar documents.
Financial Times reporting has documented recruiters receiving large volumes of AI-polished and keyword-optimised CVs.8 WIRED has since described how low-friction application technology is contributing to a sea of applications and a corresponding expansion of automated screening.9 These reports are market observations, not proof that AI alone caused the increase.
Our interpretation: AI makes the CV better as a piece of marketing and worse as a source of information.
Professional profiles are not records. They are arguments.
Empirical evidenceA 2026 NBER working paper followed monthly snapshots of 29.4 million US LinkedIn profiles. It found that 19.7% of established users had retroactively edited the title or description of a job they had already left. The typical change occurred more than four years later.5
After ChatGPT's release, retroactive additions of terms including AI, GPT and LLM rose more than sixfold. The authors estimate that a 2026 snapshot would overstate the prevalence of AI-related skills in 2022 by approximately 30%.
This is not evidence that every edit is dishonest. People may be adding detail, correcting weak descriptions or translating old work into current language. The more important point is structural: a professional profile is mutable and shaped by contemporary market incentives.
The underlying job may be unchanged. The argument made about it is not.
Once the CV is recognised as marketing, treating it as the career database becomes difficult to defend.
The alternative is not a longer CV.
Sandery interpretationA ten-page CV would preserve more information and become unusable for humans. The historical constraint would simply reappear at the interpretation stage.
AI changes the architecture because machines can economically organise, retrieve and interpret substantially more career evidence than a human can scan in a shortlisting exercise. The underlying record can remain broad while relevance is determined dynamically for a specific purpose.
The representation no longer needs to be the database. It can be generated from the database.
- Person
- Compressed CV
- Recruiter inference
- Hiring decision
- Career evidence
- Professional intelligence
- Situational interpretation
- Human decision
The market already wants evidence rather than proxies.
LinkedIn reports that 93% of talent-acquisition professionals consider accurate skills assessment important to improving quality of hire. Its platform data associates greater use of skills-based searches with a 12% higher likelihood of what LinkedIn defines as a quality hire.2 The finding is associative, not causal, but the preference is clear.
The World Economic Forum's 2025 survey of more than 1,000 employers found that 81% expected to continue using work experience when assessing candidates between 2025 and 2030. At the same time, 48% expected to use skills assessments and 43% university-degree requirements.3
Selection research provides a stronger reason to move closer to underlying evidence. A meta-analysis spanning 180 independent samples found that structured biographical information can be highly predictive, with results depending materially on what is captured and how it is scored.6 Updated meta-analytic work has also revised down several long-accepted validity estimates while retaining structured, job-relevant methods among the strongest predictors available.7
None of this implies that a database can perfectly predict performance. It implies that six selectively written bullets are an implausible optimum.
Talent-acquisition professionals who say accurate skills assessment is important to improving quality of hire.
LinkedIn, 2025Employers expecting to use skills assessments in hiring during 2025–2030.
World Economic Forum, 2025Criterion-correlation samples examined in a modern meta-analysis of employment biodata.
Speer et al., 2022CV-based filters make capable people disappear.
Harvard Business School and Accenture studied 8,720 “hidden workers” and 2,275 executives across the US, UK and Germany. Their research describes hiring systems that screen out people with employment gaps, missing credentials or terminology that does not exactly match stated criteria, including veterans whose experience is expressed in language employers do not recognise.4
The underlying problem is semantic. The person may possess the capability. The document does not express it in the expected vocabulary.
Language models can reason across related descriptions rather than requiring exact lexical matches. But semantic matching is only as useful as the evidence available to interpret. A system cannot identify a transferable capability that was discarded before the analysis began.
The people most disadvantaged by the CV are often those whose capability requires explanation.
The comparison should be evidence against evidence.
A richer system would begin with a longitudinal career record: projects completed, responsibilities held, decisions made, problems solved, transactions executed, teams managed, environments navigated, outcomes achieved and capabilities demonstrated.
The opportunity should receive the same treatment. A job description is not necessarily the role. The underlying requirement can be decomposed into the problems, decisions, knowledge, relationships and operating situations that will determine success.
The matching question then changes from whether a CV resembles the expected candidate to what evidence exists that this individual could perform the work required.
the expected profile?
could perform the work?
AI does not solve hiring. It can stop us discarding so much before it begins.
Human performance remains contextual. Capability may not transfer cleanly between environments. Motivation changes. Managers, teams, incentives and culture matter. Even the strongest selection methods predict outcomes; they do not determine them.
The objective should not be an algorithm that announces who should be hired. It should be a better information system for human-capital decisions.
Our view: the CV will probably survive in the near term because institutions, systems and users understand it. But its function can change. It can become a presentation layer generated for a specific purpose from a richer underlying record.
The same evidence could support an investment-role CV, board biography, promotion case, executive-search profile, interview evidence pack or succession analysis without any of those outputs becoming the source of truth.
The Sandery view
The future is not CV-free. It is CV-subordinate.
Sandery is building the professional record behind the document. The Career Bank is intended to preserve structured, contextual and longitudinal evidence about what an individual has actually done, rather than repeatedly reconstructing that person from a profile prepared for the last transaction.
CV Architect is being developed as the first integrated application of that record: a role-specific professional argument created from a richer source record and reviewed against the underlying facts. Guided CV work is available today through the founding-client programme.
Over time, the same intelligence can support better representation, matching, assessment and professional decisions. The document becomes an output. The career record becomes infrastructure.
The implication
The CV was an ingenious solution to an information constraint.
AI is removing much of the constraint.
The future of human-capital intelligence is unlikely to be a better CV. It is a world in which the CV is no longer where the intelligence lives.
Build the underlying record
Start before the next application.
References
Evidence used in this note.
The proposed shift from CVs as canonical records to generated presentation layers is Sandery interpretation. The evidence supports the weaknesses of proxy-based hiring, the mutability of professional profiles and growing interest in skills evidence; it does not establish that CVs will disappear on a fixed timetable.
Research & institutional evidence
- A Skills-First Labour MarketOECDEvidence on hiring proxies, skills assessment and the difficulty of validating capability, 2026 ↗
- The Future of Recruiting 2025LinkedIn Talent SolutionsPlatform research on skills assessment, skills-based search and quality of hire ↗
- Future of Jobs Report 2025World Economic ForumSurvey of more than 1,000 employers representing over 14 million workers ↗
- Hidden Workers: Untapped TalentHarvard Business School and AccentureStudy of 8,720 workers and 2,275 executives across the US, UK and Germany, 2021 ↗
- Time Travel on Professional ProfilesNicholas Bloom, Gideon Moore, Lisa K. Simon and Caelan Wilkie-RogersNBER Working Paper 35546 · Monthly snapshots of 29.4 million US LinkedIn profiles, 2026 ↗
- Meta-Analysis of Biodata in Employment SettingsAndrew B. Speer et al.Personnel Psychology · 180 independent samples of criterion correlations, 2022 ↗
- Revisiting Meta-Analytic Estimates of Validity in Personnel SelectionPaul R. Sackett et al.Journal of Applied Psychology · Revised validity estimates across selection methods, 2022 ↗